About
I am a postdoctoral researcher in the Deep Learning Theory Team at RIKEN AIP.
My research develops statistical theory and methodology for modern data analysis, with an emphasis on Bayesian statistics, spatio-temporal modeling, functional data analysis, and learning theory for large-scale models. I am especially interested in how Bayesian and decision-theoretic perspectives can clarify the behavior of overparameterized models, in-context learning, and test-time adaptation.
Before joining RIKEN AIP, I received my Ph.D. in Economics from the University of Tokyo in 2025. My dissertation was titled Nonparametric Bayesian Statistics for High-dimensional Data.
Selected Papers
View All →The Geometry of Statistical Feature Learning in Mean-Field Langevin Dynamics
Zong Shang, Tomoya Wakayama, Guillaume Lecué, Taiji Suzuki
A geometric formulation of statistical feature learning for supervised regression through mean-field Langevin dynamics.
A Decision-Theoretic View of Test-Time Training: When, How Far, and Which Directions to Adapt
Tomoya Wakayama
Proceedings of the 43rd International Conference on Machine Learning (ICML 2026)
A decision-theoretic analysis of test-time training, including adaptation distance and direction selection.
In-Context Learning Is Provably Bayesian Inference: A Generalization Theory for Meta-Learning
Tomoya Wakayama, Taiji Suzuki
Proceedings of the 43rd International Conference on Machine Learning (ICML 2026)
A generalization theory showing when in-context learning implements Bayesian inference in meta-learning.
Ensemble Prediction via Covariate-dependent Stacking
Tomoya Wakayama, Shonosuke Sugasawa
Statistics and Computing
A covariate-dependent stacking framework for ensemble prediction with oracle-type theoretical guarantees.
Similarity-based Random Partition Distribution for Clustering Functional Data
Tomoya Wakayama, Shonosuke Sugasawa, Genya Kobayashi
Journal of the Royal Statistical Society, Series C
A nonparametric Bayesian clustering method for functional data that incorporates similarity information.
Spatiotemporal Factor Models for Functional Data with Application to Population Map Forecast
Tomoya Wakayama, Shonosuke Sugasawa
Spatial Statistics
A Bayesian spatio-temporal factor model for functional data with an application to population flow forecasting.
News
Two papers were accepted to ICML 2026.
Started JSPS KAKENHI Early-Career Scientists and a RIKEN Incentive Research Project.
Named an Excellent Presentation Award Finalist at IBIS2025.
Similarity-based Random Partition Distribution for Clustering Functional Data was published in the Journal of the Royal Statistical Society, Series C.